AI Agents Aid Scientists in Complex Data Analysis

Scientists at European XFEL face significant challenges in analyzing the vast and intricate datasets generated by their experiments. This difficulty stems from the need to combine deep domain expertise with specialized knowledge about the facility and its software, which is often dispersed across various documentation, tools, and support channels. To alleviate this data analysis bottleneck, a research team developed and assessed an agentic AI system specifically tailored to the needs of these scientists. The design science research approach involved a comprehensive literature review, evaluation of 16 AI tools, interviews, a focus group, and a user study with XFEL experts. This process led to the development and evaluation of two prototypes. The study successfully identified key knowledge management challenges in scientific data analysis and derived specific requirements for an AI agent capable of assisting with knowledge retrieval and source code generation. The findings offer valuable design recommendations for creating maintainable AI support systems within highly specialized scientific environments, adaptable to the evolving landscape of AI tools.